On Wed, Sep 07, 2022 at 04:58:49PM +0530, Ravi Bangoria wrote:
> > -static void
> > -ctx_flexible_sched_in(struct perf_event_context *ctx,
> > - struct perf_cpu_context *cpuctx)
> > +/* XXX .busy thingy from Peter's patch */
> > +static void ctx_flexible_sched_in(struct perf_event_context *ctx, struct pmu *pmu)
>
> This one turned out to be very easy. Given that, we iterate over each
> pmu, we can just return error if we fail to schedule any flexible event.
> (It wouldn't be straight forward like this if we needed to implement
> pmu=NULL optimization.)
>
> ---
> diff --git a/kernel/events/core.c b/kernel/events/core.c
> index e0232e0bb74e..923656af73fe 100644
> --- a/kernel/events/core.c
> +++ b/kernel/events/core.c
> @@ -3751,6 +3751,7 @@ static int merge_sched_in(struct perf_event *event, void *data)
> cpc = this_cpu_ptr(event->pmu_ctx->pmu->cpu_pmu_context);
> perf_mux_hrtimer_restart(cpc);
> group_update_userpage(event);
> + return -EBUSY;
> }
> }
>
I'm afraid this breaks things; consider:
f79256532682 ("perf/core: fix userpage->time_enabled of inactive events")
I totally hate this -- because it means we *HAVE* to iterate the
inactive events, but alas.
